AWS Builder Center
DocuMind AI: A Full-Stack RAG Document Assistant

DocuMind AI: A Full-Stack RAG Document Assistant

I built DocuMind AI, a full-stack Retrieval-Augmented Generation (RAG) platform on AWS using FastAPI, React, and ChromaDB to chat with PDFs without hallucinations

Aspiring Cloud Engineer | Full-Stack Developer
The Problem: Reading through dense, multi-page PDFs to find specific information is a massive time sink for students and professionals. I wanted a way to instantly chat with my documents and extract exact answers without hallucination.
The Solution: I built DocuMind AI, a full-stack Retrieval-Augmented Generation (RAG) platform. Users can upload any PDF, and the app uses ChromaDB for semantic vector search to retrieve the most relevant chunks of text. It then feeds that context to the Google Gemini LLM to generate precise, markdown-formatted answers.
How I Built It (The Architecture):
  • Frontend: React and Vite (deployed as static assets)
  • Backend: Python and FastAPI
  • Vector Store: ChromaDB with Google Generative AI Embeddings
  • Hosting: AWS EC2 (Ubuntu) running an Nginx reverse proxy and a Systemd background service to keep the API alive.
How My AI Agent Helped Me Ship (Required): My AI coding agent was my dedicated DevOps engineer for this sprint. It helped me:
  1. Configure a 2GB Linux swap file to prevent out-of-memory crashes on my EC2 instance while compiling heavy Python ML packages.
  1. Write the exact Nginx reverse proxy configuration to seamlessly route traffic between my static React frontend and my FastAPI backend on port 8000.
  1. Refactor my initial keyword-search logic into true cosine similarity vector search using ChromaDB.
Live Project Link:  http://51.21.221.215/ 
Any opinions in this article are those of the individual author and may not reflect the opinions of AWS.
Enjoyed reading this content? Let the author know!

Your likes, comments, shares, and saves help creators reach more builders.

Loading recommendations

Loading article